70,000 Live Titles a Week: Why Streamers Say Algorithms Alone Fail Viewers

Fubo served customers 70,000 live titles in a single week, plus another 40,000 on-demand assets — a volume its senior product executive calls the streaming challenge that keeps him up at night. "In the past week, we've had 70,000 live titles available to our customers, plus the additional 40,000 [VOD assets], which is a lot more than any of the SVOD services," said Isaac Josephson, senior vice president of product management at Fubo, speaking at NAB Show NY this week.

Josephson appeared on a streaming panel alongside Srebrina Tanova, vice president of product management for tech, enterprise content at Skydance, and Steve Leardi, vice president of product design at Peacock. Their shared theme: personalization has hard limits, and the industry's default tools — recommendation algorithms and onboarding questionnaires — often fail the viewer.

Why do recommendation algorithms fail at content discovery?

Algorithms carry a built-in popularity bias that narrows what viewers see, according to Skydance's Tanova. The bias compounds over time and actively damages discovery rather than helping it.

"The way personalization models work is they have a sway toward popularity bias, and they inherently over time become biased systems. So what happens is you just start seeing a homepage where it's mostly just the most popular content starting to show in a very repetitive form, and there's very little diversity [of content]," Tanova said. "Now, obviously, there are algorithmic things that you can do to address that, but ultimately, these models have bias, so it actually starts being a detriment to content discovery."

The problem intensifies when a platform's catalog mixes live sports, live events, on-demand entertainment and news. "How can you anticipate their needs, and how do you not overwhelm the customer? I think that's a common theme that a lot of streaming platforms struggle with when you're dealing with such a vast volume of content that is so varied," Tanova said.

Fubo's answer is a layered approach rather than pure automation. Personalization based on a user's past behavior is not enough, Josephson said. "You have to mix it up with segmentation-level personalization and editorial curation to make sure that your users are getting all of the value that they can out of your experience," he said.

Segmentation-level personalization means tailoring the experience to broad audience groups, not single users. Editorial curation means human programmers hand-pick and arrange content.

Do onboarding profiles actually work?

Mostly, no. Platforms routinely ask new customers to complete an onboarding flow or create a viewer profile. Tanova said a massive percentage of users skip the step — and she counts herself among them.

The skipped step creates a data problem. "Users are just trying to watch content, so they're going to pick whatever if they have to make a selection. Now that information that they input just to try to get out of the onboarding goes into their personalized experience, and they're now going into an experience that's showing them stuff they don't care about because they were just trying to bypass the onboarding pop-up," she said.

Bad input data produces a bad personalized homepage. The mechanism is simple: the platform learns from choices the user never seriously made.

What do viewers mean when they ask for personalization?

Peacock's Leardi argued customers asking for personalization are usually asking for something else: immersion. "It's not necessarily finding the title that is next to this or adjacent to this that fits in the same genre. It's about 'I love this show, and I want to go deeper and be completely consumed by this,'" Leardi said.

That principle drives the NBCUniversal-owned streamer's new mobile feature, "Your Bravoverse." The TikTok-style vertical feed serves snackable clips from the Bravo library, tailored to each user, with an AI avatar of host Andy Cohen guiding viewers through storylines.

Superfans use it to go deeper into franchises they already love. But the feed also functions as a discovery engine. Users tap a clip and go straight to the full episode. "They are using it to find something to watch," Leardi said. "[It's] creating that easy pathway directly to a long-form asset, and it's just a new form of discovery that is very mobile centric."

What did Peacock learn from its NBA altcasts?

Peacock launched NBA altcasts (alternative broadcast feeds offering a different camera angle or presentation of the same game) this year, including a "Courtside Live" feature for the 2025-26 season. The first iteration produced a concrete lesson in what viewers actually want.

"We installed two robo cameras on the scorer's table pointed at the bench. What we found is we got a lot of shots of coaches' backsides, [which is] not exactly what customers wanted," Leardi said. "Ahead of this season, we had this realization when people meant they wanted to get closer to the game, what they really meant is, 'I want to get closer to the athletes.'"

Peacock responded with two star cams, each isolated on a specific player — on the bench, cheering teammates or reacting when the team blows a lead.

The star cams also address an engagement demand Leardi hears from younger viewers. "How are they engaging? That's something, especially from a younger audience, we've heard a lot about. They just really want to find that connection. So we're always learning," he said.

The panel's collective message for marketers and product teams: automated personalization degrades without human curation layered on top, and the winning discovery formats — vertical feeds, player-isolated cameras — are the ones built around connection rather than adjacency.